Finding gene coherent patterns using PATSUB+

Abstract

This paper presents a pattern aided subspace clustering algorithm (PATSUB+) to find biologically relevant coexpressed patterns in gene expression data. The algorithm works iteratively over a relative expression domain using NMRS (Normalized Mean Residue Similarity) measure, it discovers a set of possibly overlapping subspace clusters for any gene expression matrix. PATSUB+ performs satisfactorily while tested on several real life gene expression datasets and compared with several competing algorithms.